多个gpu训练，卡住不动，gpu占用率为0

[lvpchen]

多个gpu训练，一直卡在下面，请问下怎么回事？

I1021 16:08:19.171383 4102 solver.cpp:70] Solver scaffolding done.
I1021 16:08:19.176095 4102 caffe.cpp:255] Starting Optimization
I1021 16:08:19.751611 4243 upgrade_proto.cpp:79] Attempting to upgrade batch norm layers using deprecated params: mobilenet_yolov3_lite_train.prototxt
I1021 16:08:19.751641 4243 upgrade_proto.cpp:82] Successfully upgraded batch norm layers using deprecated params.
I1021 16:08:19.755781 4243 annotated_data_layer.cpp:79] output data size: 1,3,608,608
I1021 16:08:19.823746 4243 upgrade_proto.cpp:79] Attempting to upgrade batch norm layers using deprecated params: mobilenet_yolov3_lite_test.prototxt
I1021 16:08:19.823776 4243 upgrade_proto.cpp:82] Successfully upgraded batch norm layers using deprecated params.
I1021 16:08:19.823781 4243 solver.cpp:203] Creating test net (#0) specified by test_net file: mobilenet_yolov3_lite_test.prototxt

[lvpchen]

@eric612

[kingmacth]

1. 修改CMakeLists.txt文件第33行将OFF修改ON 开启多GPU支持
2. 修改src\caffe\solver.cpp 第70行增加InitTestNets(); 解决多GPU卡顿